A Lightweight Neural Model for Biomedical Entity Linking

Abstract

Biomedical entity linking aims to map biomedical mentions, such as diseases and drugs, to standard entities in a given knowledge base. The specific challenge in this context is that the same biomedical entity can have a wide range of names, including synonyms, morphological variations, and names with different word orderings. Recently, BERT-based methods have advanced the state-of-the-art by allowing for rich representations of word sequences. However, they often have hundreds of millions of parameters and require heavy computing resources, which limits their applications in resource-limited scenarios. Here, we propose a lightweight neural method for biomedical entity linking, which needs just a fraction of the parameters of a BERT model and much less computing resources. Our method uses a simple alignment layer with attention mechanisms to capture the variations between mention and entity names. Yet, we show that our model is competitive with previous work on standard evaluation benchmarks.

Knowledge-graph-enabled biomedical entity linking: a survey

“…However, efficiency and scalability are essential for real-time and large-scale applications. Although Lai et al [84] and Chen et al [86] aimed to determine how to improve the efficiency of EL tasks, they did not perform tests on large datasets. Therefore, an essential direction for future research is to investigate the design of systems that substantially improve the efficiency and scalability of BM-EL systems while maintaining high accuracy and precision.…”

“…Chen et al [86] also used CNN as the backbone network of the model to significantly reduce the model parameters. They further introduced more features into the embedding representation of mention and entity to improve the model performance.…”

“…In each report, the disorder mention is linked to the corresponding entity in the SNOMED-CT knowledge base. If no corresponding entity exists, it is labeled as "CUI-less" (about 28.2% in the training set and 32.7% in the test set [86]). The dataset is separated into training (199) and test (99) subsets [23].…”

“…Zhao et al[36] achieves the best results on the BC5CDR dataset and owns the leadership on the NCBI dataset, showing that the joint modeling of medical named-entity recognition and normalization has advantages over pipeline implementation owing to their mutually beneficial relationship. Owing to the selection of the backbone network and the introduction of more high-quality features, Chen et al[86], Lai et al[84] maintain a high level of model performance despite the significant improvement in model inference speed and reduction in model size, which can bring comparative advantages to the system deployment and application.…”

Multilingual bi‐encoder models for biomedical entity linking

“…Other biomedical studies can be found for EL. Chen, Wang, et al (2020) and Chen, Varoquaux, et al (2020) proposed a lightweight neural method for biomedical EL. The method only needs some of the parameters of the BERT model.…”